Where the figures come from, and what they do and do not mean.

Where the data comes from

There are two kinds of source. Each record keeps a reference to the page or document it came from.

• The public WKF ranking pages. Athlete profiles, ranking positions and points, and the lists of events and results that those pages show. A ranking observation is dated by when it was retrieved, so a change over time can only be shown from the observations that were actually stored.
• Tournament result documents (PDF). The source of match-level results: who met whom, in which round, with what points, and the kata performed. These are read by software and then reviewed by a person (next section).

The two are kept apart. A row read from a source is evidence, not a fact the application vouches for, and sources that disagree are recorded as a conflict with both values kept.

How tournament results are reviewed and approved

A tournament document does not count until a person has approved it. The steps:

1. Read. Software extracts the event, categories, participants, bracket and results from the PDF. This is a proposal, not a record.
2. Review and correct. A person compares the proposal with the source document and corrects it. Every edit is kept with its before and after value, who made it and when.
3. Validate. The server checks the bracket for contradictions, for example the same participant in both slots, a winner with fewer points than the loser, or a progression that points nowhere. Errors block approval. Warnings, such as a match with no established winner, are shown for the reviewer to weigh.
4. Approve or reject. Approval writes everything in one step or not at all, and records who approved which version. Rejection writes nothing to the competition data.
5. Link participants. A participant counts towards an athlete's history only once linked to that athlete. Unlinked participants appear in nobody's figures, and ambiguous cases stay for a person to decide.

A mistake found after approval is fixed by a correction: the approved data stays published while a working copy is edited, and the result is stored as a new revision with every change listed. Nothing is overwritten silently, and importing the same document again changes nothing.

How to read a figure

The analytics are descriptive. They count what the approved records say. They are not a ranking, a rating or a prediction, and they do not say who is the better athlete.

What is missing

These are the known limits of the data today. They are stated so that a missing figure is not read as a finding.

• Match-level results exist only for tournaments whose documents have been approved. Not every athlete or event is included, and the athletes held are the ones that were retrieved, not everyone who competes.
• A WKF results page lists a category's top places, not every competitor. A row with no place is not a last place.
• The tournament sheets seen do not record where a losing competitor goes next, so entry into repechage is not recorded. A repechage match is analysed because it is printed under a repechage heading, not because of an earlier loss.
• Judges' scores are not read from the documents, and some matches have no points recorded, so averages use fewer matches and say how many.
• Kata text that cannot be matched to the official numbered list is kept as printed and left out of the repertoire count.
• Some documents carry no date, and ranking data is as current as its last retrieval.